mams / README.md
xinxngxin's picture
Add researched SOTA result
292cd37
|
raw
history blame
No virus
2.55 kB
> 上述数据集为ABSA(Aspect-Based Sentiment Analysis)领域数据集,基本形式为从句子中抽取:方面术语、方面类别(术语类别)、术语在上下文中情感极性以及针对该术语的观点词,不同数据集抽取不同的信息,这点在jsonl文件的“instruction”键中有分别提到,在此我将其改造为了生成任务,需要模型按照一定格式生成抽取结果。
#### 以acos数据集中抽取的jsonl文件一条数据举例:
```
{
"task_type": "generation",
"dataset": "acos",
"input": ["the computer has difficulty switching between tablet and computer ."],
"output": "[['computer', 'laptop usability', 'negative', 'difficulty']]",
"situation": "none",
"label": "",
"extra": "",
"instruction": "
Task: Extracting aspect terms and their corresponding aspect categories, sentiment polarities, and opinion words.
Input: A sentence
Output: A list of 4-tuples, where each tuple contains the extracted aspect term, its aspect category, sentiment polarity, and opinion words (if any). Supplement: \"Null\" means that there is no occurrence in the sentence.
Example:
Sentence: \"Also it's not a true SSD drive in there but eMMC, which makes a difference.\"
Output: [['SSD drive', 'hard_disc operation_performance', 'negative', 'NULL']]'
"
}
```
> 此处未设置label和extra,在instruction中以如上所示的字符串模板,并给出一个例子进行one-shot,ABSA领域数据集(absa-quad,acos,arts,aste-data-v2,mams,semeval-2014,semeval-2015,semeval-2016,towe)每个数据集对应instruction模板相同,内容有细微不同,且部分数据集存在同一数据集不同数据instruction内容不同的情况。
#### 原始数据集
- 数据[链接](https://github.com/siat-nlp/MAMS-for-ABSA)
- Paper:[A Challenge Dataset and Effective Models for Aspect-Based Sentiment Analysis](https://aclanthology.org/D19-1654.pdf)
- 说明:原始数据由MAMS-ACSA和MAMS-ATSA组成,两部分数据集为不同任务,抽取不同元素。
#### 当前SOTA
*数据来自[PaperWithCode](https://paperswithcode.com/sota/aspect-based-sentiment-analysis-on-mams)*
- 评价指标:Accuracy 、 Macro-F1
- 模型:RGAT+ (Accuracy: **84.52** , Macro-F1: **83.74**)
- Paper:[Investigating Typed Syntactic Dependencies for Targeted Sentiment Classification Using Graph Attention Neural Network](https://paperswithcode.com/paper/exploiting-typed-syntactic-dependencies-for)